There's something Knuth is not addressing. If I understand Dijkstra correctly he's seeing an enormous detour that we're taking with, for example, languages whic

There's something Knuth is not addressing. If I understand Dijkstra correctly he's seeing an enormous detour that we're taking with, for example, languages which evolved from a computer architecture. These are not formal, they're for the most part convenience layers and leaky abstractions atop a machine. Compare this to other disciplines. Imagine a modern bridge builder looking at the bridges of old, and come up with a model just large enough to communicate to other bridge builders how bridges should be made. If the bridge builders of old "just did something", what is our modern bridge builder modelling then? You could argue, "but those bridges are still around, so they probably did something well". Sure, but that's just settling for a local optimum. If I want a good modern bridge, I hope that engineer is taking physics into account -- and not spending their time on how the old bridge builders were throwing with mud and stones. We can't stand on shoulders if we keep on regurgitating spoiled food.